Notebook: Make llm.c thunder-compatible and use native llm.c CUDA kernels

[nikitaved]

As per title!

We kick-off with a PyTorch reference model, then replace some of its parts with the llm.c's CUDA kernels. We also test the outcome against the C implementation (using provided tools) and identify potential future work.

Maybe we want to actually provide C-Python bindings for tightener benchmarks?

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[lantiga]

@nikitaved should we merge this one so we keep it as a reference? Maybe let's add a WIP caveat, demonstration purposes only.

[lantiga]

tagging older Draft PR's at later for tidiness

@nikitaved feel free to keep chipping at this or reopen if this gets back being active